Arvidsson, R. and McKelvey, T., "Battery Parameter Estimation from Recorded Fleet Data," SAE Technical Paper 2016-01-2360, 2016, doi:10.4271/2016-01-2360.
Existing battery parameter model structures are evaluated by estimating model parameters on real driving data applying standard system identification methods. Models are then evaluated on the test data in terms of goodness of fit and RMSE in voltage predictions. This is different from previous battery model evaluations where a common approach is to train parameters using standardized tests, e.g. hybrid pulse-power capability (HPPC), with predetermined charge and discharge sequences. Equivalent linear circuit models of different complexity were tested and evaluated in order to identify parameter dependencies at different state of charge levels and temperatures. Models are then used to create voltage output given a current, state of charge and temperature. The average accuracy of modelling the DC bus voltage provides a model goodness of fit average higher than 90% for a single RC circuit model. Both single RC equivalent circuit model and R-equivalent circuit model produce goodness of fit at average 75 % or higher. The dual RC equivalent circuit model experienced larger errors in voltage estimations compared to single R and RC equivalent circuit models.